Transferring Procedural Knowledge across Commonsense Tasks

Stories about everyday situations are an essential part of human communication, motivating the need to develop AI agents that can reliably understand these stories. Despite the long list of supervised methods for story completion and procedural understanding, current AI has no mechanisms to automatically track and explain procedures in unseen stories. To bridge this gap, we study the ability of AI models to transfer procedural knowledge to novel narrative tasks in a transparent manner. We design LEAP: a comprehensive framework that integrates state-of-the-art modeling architectures, training regimes, and augmentation strategies based on both natural and synthetic stories. To address the lack of densely annotated training data, we devise a robust automatic labeler based on few-shot prompting to enhance the augmented data. Our experiments with in- and out-of-domain tasks reveal insights into the interplay of different architectures, training regimes, and augmentation strategies. LEAP's labeler has a clear positive impact on out-of-domain datasets, while the resulting dense annotation provides native explainability.

[1]  E. Hovy,et al.  A Survey of Active Learning for Natural Language Processing , 2022, EMNLP.

[2]  Graham Neubig,et al.  Language Models of Code are Few-Shot Commonsense Learners , 2022, EMNLP.

[3]  Eric Nyberg,et al.  Coalescing Global and Local Information for Procedural Text Understanding , 2022, COLING.

[4]  Nancy Fulda,et al.  Enhanced Story Comprehension for Large Language Models through Dynamic Document-Based Knowledge Graphs , 2022, AAAI.

[5]  J. Pujara,et al.  Understanding Narratives through Dimensions of Analogy , 2022, ArXiv.

[6]  Mo Yu,et al.  A Survey of Machine Narrative Reading Comprehension Assessments , 2022, IJCAI.

[7]  Marc-Alexandre Côté,et al.  ScienceWorld: Is your Agent Smarter than a 5th Grader? , 2022, EMNLP.

[8]  M. Lewis,et al.  Rethinking the Role of Demonstrations: What Makes In-Context Learning Work? , 2022, Conference on Empirical Methods in Natural Language Processing.

[9]  H. V. Hoof,et al.  Leveraging class abstraction for commonsense reinforcement learning via residual policy gradient methods , 2022, IJCAI.

[10]  Quoc V. Le,et al.  Finetuned Language Models Are Zero-Shot Learners , 2021, ICLR.

[11]  Weizhu Chen,et al.  What Makes Good In-Context Examples for GPT-3? , 2021, DEELIO.

[12]  J. Chai,et al.  Tiered Reasoning for Intuitive Physics: Toward Verifiable Commonsense Language Understanding , 2021, EMNLP.

[13]  Wojciech Zaremba,et al.  Evaluating Large Language Models Trained on Code , 2021, ArXiv.

[14]  Tommaso Pasini,et al.  ESC: Redesigning WSD with Extractive Sense Comprehension , 2021, NAACL.

[15]  Parisa Kordjamshidi,et al.  Time-Stamped Language Model: Teaching Language Models to Understand The Flow of Events , 2021, NAACL.

[16]  Laria Reynolds,et al.  Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm , 2021, CHI Extended Abstracts.

[17]  Pedro A. Szekely,et al.  CSKG: The CommonSense Knowledge Graph , 2020, ESWC.

[18]  Yonatan Bisk,et al.  Knowledge-driven Data Construction for Zero-shot Evaluation in Commonsense Question Answering , 2020, AAAI.

[19]  Tao Qin,et al.  Knowledge-Aware Procedural Text Understanding with Multi-Stage Training , 2020, WWW.

[20]  Filip Ilievski Story Generation with Commonsense Knowledge Graphs and Axioms , 2021 .

[21]  N. Mostafazadeh,et al.  GLUCOSE: GeneraLized and COntextualized Story Explanations , 2020, Conference on Empirical Methods in Natural Language Processing.

[22]  Mark Chen,et al.  Language Models are Few-Shot Learners , 2020, NeurIPS.

[23]  Wanxiang Che,et al.  Discriminative Sentence Modeling for Story Ending Prediction , 2019, AAAI.

[24]  Yejin Choi,et al.  PIQA: Reasoning about Physical Commonsense in Natural Language , 2019, AAAI.

[25]  Doug Downey,et al.  Abductive Commonsense Reasoning , 2019, ICLR.

[26]  Omer Levy,et al.  RoBERTa: A Robustly Optimized BERT Pretraining Approach , 2019, ArXiv.

[27]  Ting Liu,et al.  Story Ending Prediction by Transferable BERT , 2019, IJCAI.

[28]  Aditya Gupta,et al.  Tracking Discrete and Continuous Entity State for Process Understanding , 2019, SPNLP@NAACL-HLT.

[29]  Zhou Yu,et al.  Incorporating Structured Commonsense Knowledge in Story Completion , 2018, AAAI.

[30]  Bhavana Dalvi,et al.  Tracking State Changes in Procedural Text: a Challenge Dataset and Models for Process Paragraph Comprehension , 2018, NAACL.

[31]  Chris Dyer,et al.  The NarrativeQA Reading Comprehension Challenge , 2017, TACL.

[32]  Omer Levy,et al.  Simulating Action Dynamics with Neural Process Networks , 2017, ICLR.

[33]  Jerry R. Hobbs,et al.  A Formal Theory of Commonsense Psychology: How People Think People Think , 2017 .

[34]  Choh Man Teng,et al.  Broad Coverage, Domain-Generic Deep Semantic Parsing , 2017, AAAI Spring Symposia.

[35]  George A. Miller,et al.  Introduction to WordNet: An On-line Lexical Database , 1990 .

[36]  R. Schank,et al.  Scripts, plans, and knowledge , 1975, IJCAI 1975.

[37]  Eugene Charniak,et al.  Toward a model of children's story comprehension , 1972 .